Neuron Explorer

Qwen3-0.6B

comma that ends a clause before a result

A neuron in layer 18 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featured, so it has steering outputs.

What this neuron does

This neuron fires on a comma that finishes a full statement and sets up a consequence or extra detail, such as 'associated with a lower risk of death from any cause, finds...'. Turning it up made the model produce filler phrases like 'for example' over and over.

Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).

Fires most on

The text windows where this neuron's activation was highest. Amber marks positive activation; the outlined token is the peak. Hover a token for its strength.

  1. #1 peak

    particularly plant protein, linked to lower risk of death⏎Diets high in protein, particularly plant protein, are associated with a lower risk of death from any cause, finds an analysis of the latest evidence published by The BMJ today. ...continue

  2. #2 peak

    a kind of antioxidant are believed to play a significant role in whittling your middle.⏎Lab rats who were fed a blueberry-rich mixture had less abdominal fat, lower triglycerides, lower cholesterol, and improved fasting glucose and insulin sensitivity which

  3. #3 peak

    but once you see the incredible benefits you could never look back. It cooks your food and keeps it piping hot. The more you incorporate it into your daily life, the more you will find you use it.⏎A degree of experimentation is necessary to

  4. #4 peak

    ews⏎The use of active voice is important while writing online content. It improves the readability of your copy and reduces the wordiness. The lines seem more direct, so this is a good way to avoid fluff.⏎Active and Passive Voice –

  5. #5 peak

    reasons. … First and foremost, plastic is much lighter and more aerodynamic than metal, which improves the car’s fuel efficiency. Plastic is also easier to shape, which is

  6. #6 peak

    body’s primary energy source. Your body cannot absorb the disaccharide, or two-sugar molecule, as is, so it must first sever the chemical

Steering

Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.

turned down a lot
Prompt

Here is something I have been thinking about lately: for a while, I have been in the group of both of me, for a short period of time, for the rest of the time, I am, in the end, the sum of both of me, for a short period of

Most negative on

Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.

  1. #1 peak

    ($) before the answer.⏎Subtracting money is easy, too, so long as it is once more remembered that money has to be written and subtracted the way that decimals are written and subtracted. Dollars go to the left of the decimal

  2. #2 peak

    15 S.W. Second St. in Corvallis. It is free and open to the public. In November, Science Pub will resume at its usual location at the Old World Deli.⏎

  3. #3 peak

    sponsored by the Jewish Student Union and supported by Campus Ambassadors, Students Against Slavery and Willamette University's Religious Studies Department.⏎For more information contact Jessica Sternbach at firstname.lastname@example.org.<|endoftext|>St. Peter's Cathedral is one of

  4. #4 peak

    son knew this day was coming. Of the hundreds of thousands of people left stranded last week on foreign soil—hostages to an ash-spewing hell vent whose name only Icelanders could pronounce—the B.C. volcanologist might have been best equipped

Connections

Show on map

Which neurons in earlier layers feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.

Fed by earlier layers

  1. L10:556 suppresses
  2. L10:1278 excites
  3. L9:74 suppresses
  4. L13:44 excites inside web addresses and citations
  5. L7:92 suppresses
  6. L11:223 suppresses
  7. L9:207 excites
  8. L15:184 suppresses
  9. L4:42 excites
  10. L10:873 suppresses
  11. L8:328 weakly suppresses
  12. L4:92 weakly excites
  13. L1:2790 weakly suppresses
  14. L10:155 weakly suppresses
  15. L2:166 weakly excites
  16. L10:370 weakly excites
  17. L12:35 weakly suppresses
  18. L4:236 weakly suppresses
  19. L10:565 weakly suppresses
  20. L0:509 weakly suppresses
  21. L9:2581 weakly suppresses
  22. L16:1309 weakly excites
  23. L14:2375 weakly excites
  24. L10:0 weakly suppresses

Feeds later layers

  1. L21:2512 excites
  2. L19:2999 excites
  3. L19:1265 excites
  4. L19:1409 weakly excites
  5. L25:1337 weakly suppresses
  6. L24:2331 weakly suppresses
  7. L20:2132 weakly excites "the" starting an object noun phrase
  8. L20:2038 weakly excites
  9. L22:2055 weakly excites
  10. L23:1540 weakly excites
  11. L26:1113 weakly suppresses
  12. L25:796 weakly suppresses
  13. L19:2302 weakly suppresses
  14. L19:298 weakly excites
  15. L19:1873 weakly excites
  16. L19:693 weakly excites
  17. L23:1086 weakly suppresses
  18. L22:281 weakly suppresses
  19. L23:2859 weakly suppresses
  20. L27:2691 weakly suppresses
  21. L19:3065 weakly suppresses 'for' after a span of time
  22. L22:1497 weakly suppresses
  23. L21:2749 weakly suppresses
  24. L26:1528 weakly suppresses

Activation histogram

How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.

2 values between -7.48 and -6.993 values between -6.99 and -6.502 values between -6.50 and -6.003 values between -6.00 and -5.513 values between -5.51 and -5.018 values between -5.01 and -4.5210 values between -4.52 and -4.0323 values between -4.03 and -3.5332 values between -3.53 and -3.0445 values between -3.04 and -2.5477 values between -2.54 and -2.0593 values between -2.05 and -1.56133 values between -1.56 and -1.06229 values between -1.06 and -0.57396 values between -0.57 and -0.07643 values between -0.07 and 0.42523 values between 0.42 and 0.92377 values between 0.92 and 1.41322 values between 1.41 and 1.90267 values between 1.90 and 2.40195 values between 2.40 and 2.89159 values between 2.89 and 3.3997 values between 3.39 and 3.8859 values between 3.88 and 4.3839 values between 4.38 and 4.8745 values between 4.87 and 5.3629 values between 5.36 and 5.8614 values between 5.86 and 6.3516 values between 6.35 and 6.8514 values between 6.85 and 7.3412 values between 7.34 and 7.837 values between 7.83 and 8.338 values between 8.33 and 8.822 values between 8.82 and 9.326 values between 9.32 and 9.815 values between 9.81 and 10.313 values between 10.31 and 10.803 values between 11.29 and 11.794 values between 11.79 and 12.28offnegativepositiveoften

Logit lens

The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layers instead.

Promotes

  1. “This
  2. ammers
  3. liệu
  4. veys
  5. Aws
  6. ulations
  7. “A
  8. qualifications
  9. 等内容
  10. клад

Suppresses

  1. า
  2. Stateless
  3. 叶
  4. EMPL
  5. elen
  6. :]
  7. lessly
  8. out
  9. 州
  10. 燮